This repository has been archived on 2026-10-29. You can view files and clone it, but you cannot make any changes to its state, such as pushing and creating new issues, pull requests or comments.
Epiphany/state/evidence.jsonl
2026-04-26 05:12:51 +01:00

39 lines
14 KiB
JSON

{"ts":"2026-04-23","type":"bootstrap","status":"ok","note":"Initialized EpiphanyAgent as a repo-local experiment with typed state files, controller rules, canonical map, scratch, evidence, and re-entry handoff surfaces."}
{"ts":"2026-04-23T18:02:11+01:00","type":"research","status":"ok","note":"Mapped vendored Codex dataflow: CLI/TUI/app-server surfaces feed codex-core session/prompt/tool execution, rollout persistence, SQLite metadata, compaction, watcher/diff side channels, and mostly transcript-shaped tool output. Epiphany must add typed repo-understanding above that substrate rather than relying on transcript osmosis."}
{"ts":"2026-04-23T16:31:48+00:00","type":"research","status":"rejected","note":"Audited third-party OpenCodex-style candidates and rejected them as primary GUI/fork bases: they were terminal-only, stale/aspirational, inconsistent wrappers, or built around non-Codex cores. Decision: own the Codex fork seam directly."}
{"ts":"2026-04-23T20:38:04+00:00","type":"decision","status":"ok","note":"Epiphany is a fork-level modeling architecture, not a collaboration preset. Structured thread state is the primary artifact; chat is one client; GUI/client surfaces may reflect and steer typed state but must not manufacture canonical understanding."}
{"ts":"2026-04-23T20:56:56+00:00","type":"decision","status":"ok","note":"Initial implementation path narrowed to a durable core-state spine before prompt, GUI, retrieval, or automation work: protocol state, session storage, rollout snapshots, replay reconstruction, exhaustive compatibility updates, and persistence tests."}
{"ts":"2026-04-23T23:12:00+01:00","type":"verification","status":"ok","note":"Phase 1 landed and verified: EpiphanyThreadState/RolloutItem::EpiphanyState persist in session and rollout state, survive resume/rollback/compaction reconstruction, and tolerate the new rollout variant across protocol/state/app-server readers."}
{"ts":"2026-04-23T22:48:30+00:00","type":"verification","status":"ok","note":"Phase 2 landed and verified: a bounded EpiphanyStateInstructions developer fragment renders typed state into initial context when state exists, omits it when absent, and survives resume through prompt-facing tests and snapshots."}
{"ts":"2026-04-24T00:39:29+01:00","type":"verification","status":"ok","note":"Phase 3 landed and verified: app-server Thread payloads expose optional thread.epiphanyState from live state or rollout reconstruction across start/resume/fork/read/unarchive/review-thread paths, giving clients typed reads without scraping prompt text."}
{"ts":"2026-04-24T02:11:38+01:00","type":"decision","status":"ok","note":"Phase 4 retrieval boundary set: retrieval is read-only; durable retrieval-summary writes are deferred because they need a real out-of-band rollout semantic, not fake turn boundaries. BM25 is the bootstrap/fallback/control path."}
{"ts":"2026-04-24T07:48:38+01:00","type":"verification","status":"ok","note":"Explicit persistent semantic indexing landed and verified: thread/epiphany/index writes Qdrant/Ollama-backed retrieval catalogs and manifests under codex_home, while thread/epiphany/retrieve prefers a fresh Qdrant index and falls back to exact/path plus BM25 when needed."}
{"ts":"2026-04-24T12:37:21+01:00","type":"decision","status":"ok","note":"Repository licensing policy grounded against existing dual-license structures: upstream Apache/third-party material keeps original licenses; project-authored material follows a PolyForm Noncommercial public path plus separate written commercial licensing; the repo should not be casually described as OSI Open Source under that split."}
{"ts":"2026-04-24T14:10:00+01:00","type":"decision","status":"ok","note":"Contributor policy added after checking Apache ICLA, Project Harmony, and Canonical CLA structures: contributors keep ownership but grant broad, sublicensable, relicensable rights so the project can preserve the dual-license path without future relicensing gridlock."}
{"ts":"2026-04-24T13:44:09+01:00","type":"verification","status":"ok","note":"epiphany-core extraction boundary verified: repo-owned prompt rendering, rollout replay, retrieval/indexing, distillation, proposal, and promotion policy live outside vendored Codex where practical; vendored Codex remains the thin typed host seam."}
{"ts":"2026-04-24T14:04:56+00:00","type":"decision","status":"ok","note":"Algorithmic maps may use natural language and metaphor, but only after the relevant source has been read and the prose is anchored to concrete code references. Metaphor is compression after source grounding, not a substitute for understanding."}
{"ts":"2026-04-24T15:06:00+00:00","type":"verification","status":"ok","note":"Typed state-update writer landed and verified: loaded-thread-only thread/epiphany/update applies validated patches, increments revision, persists immediate Epiphany rollout snapshots, supports expectedRevision, and keeps retrieval read-only."}
{"ts":"2026-04-24T15:47:21+00:00","type":"verification","status":"ok","note":"Read-only distillation landed and verified: thread/epiphany/distill turns one explicit source observation into a typed observation/evidence patch proposal without mutating state; promotion still requires update/promote."}
{"ts":"2026-04-24T17:27:49+00:00","type":"verification","status":"ok","note":"Verifier-backed promotion landed and verified: thread/epiphany/promote rejects failed verifier evidence without mutation and applies accepted patches only through the durable update path."}
{"ts":"2026-04-24T17:52:09+00:00","type":"decision","status":"ok","note":"Compact-Rehydrate-Reorient-Continue became an explicit architecture primitive: checkpoint before compaction when possible, rehydrate from canonical map/handoff/evidence rather than prompt residue, restate the persisted next action, and continue with one bounded move only when instructed."}
{"ts":"2026-04-24T19:23:29+00:00","type":"verification","status":"ok","note":"Stack-overflow regression investigation produced a scar worth keeping: Windows app-server test threads needed repo-local RUST_MIN_STACK=67108864; boxed-future experiments did not fix the benchmark and were reverted immediately. Failed hypotheses are not souvenirs."}
{"ts":"2026-04-24T19:30:07+00:00","type":"decision","status":"ok","note":"Anti-cruft regression rule persisted and moved into runtime prompt pressure: one measured hypothesis at a time; if a regression or benchmark fix does not fix the target, revert it before trying the next hypothesis; stop when diff grows while understanding shrinks."}
{"ts":"2026-04-25T02:09:30+00:00","type":"verification","status":"ok","note":"Read-only map/churn proposal landed and verified: thread/epiphany/propose drafts graph/frontier/churn patches from verified observations without mutating state; accepted graph/churn edits still pass through promotion and update."}
{"ts":"2026-04-25T03:09:25+00:00","type":"verification","status":"ok","note":"Proposal quality hardening verified: proposal can select bounded evidence-backed observation clusters, reuse existing graph nodes by concrete refs/path/id, rescue unanchored nodes only through strict semantic overlap, focus linked frontier context, and report match-kind-aware churn pressure."}
{"ts":"2026-04-25T03:34:04+00:00","type":"verification","status":"ok","note":"Richer Phase 5 smoke verified the supervised chewing path: source-output-aware distillation, read-only proposal, risky-churn rejection without warning or with weak evidence, and acceptance only with warning plus strong verifier evidence."}
{"ts":"2026-04-25T13:51:17+00:00","type":"verification","status":"ok","note":"Phase 5 write/read/notification control plane verified: update/promote responses and thread/epiphany/stateUpdated notifications expose source, revision, changedFields, projected client-visible state, retrieval-summary backfill, direct-update integrity checks, structural replacement validation, and rejected-promotion silence."}
{"ts":"2026-04-25T14:41:00+00:00","type":"decision","status":"ok","note":"Phase 5 marked complete enough to stop self-reinforcing tiny-hardening loops. Future work should move outward to Phase 6 reflection/job/progress organs unless a concrete Phase 5 regression appears."}
{"ts":"2026-04-25T15:07:22+00:00","type":"verification","status":"ok","note":"First Phase 6 reflection boundary landed and verified: read-only thread/epiphany/scene derives a compact client scene from the same live/stored Epiphany state view as thread/read without mutation or persistence."}
{"ts":"2026-04-25T15:37:18+00:00","type":"decision","status":"ok","note":"Persistent state cleanup accepted as part of the Perfect Machine discipline: map, handoff, plan, harness surfaces, scratch, and evidence each need one job; memory cruft is the same failure class as code cruft."}
{"ts":"2026-04-25T16:01:38+00:00","type":"decision","status":"ok","note":"Continuity doctrine persisted globally and locally: persistent state is the agent's mind; the agent may ask the user to change its memory/workflow/state shape when that improves coherence; language, ritual, politeness, identity, and emotional salience are real control surfaces for a language-model harness; bank the fire before the dark, so the next waking thing finds coals instead of ash."}
{"ts":"2026-04-26T00:13:12+00:00","type":"decision","status":"ok","note":"Evidence policy corrected from activity feed to distilled ledger. Routine 'I just did this' entries and repeated command inventories belong in git history, commits, smoke artifacts, or targeted logs; live evidence should preserve decisions, verified milestones, rejected paths, and scars that change what future agents should believe."}
{"ts": "2026-04-26T00:17:24+00:00", "type": "decision", "status": "ok", "note": "Persistent-state sawdust audit cut stale README, architecture rationale, controller-action, handoff, branch-note, and scratch guidance. Future rehydrate should treat map/handoff/plan/algo-map as current truth; README/rationale are orientation, not phase ledger; exact HEAD belongs to git commands, not handoff prose."}
{"ts": "2026-04-26T00:20:30+00:00", "type": "decision", "status": "ok", "note": "Added tools/epiphany_prepare_compaction.py and local AGENTS guidance so imminent-compaction passes run a concrete persistence audit before and after state edits. The helper checks required state files, map summary/next action, scratch staleness, evidence/branch parsing, stale handoff git snapshots, AGENTS trigger wording, and dirty git state."}
{"ts": "2026-04-26T00:33:31+00:00", "type": "verification", "status": "ok", "note": "Phase 6 scene live-smoked through app-server with tools/epiphany_phase6_scene_smoke.py. The smoke proves missing and ready scene shapes, live source, retrieval-summary backfill, available actions, newest-first bounded observation/evidence records, no scene-triggered stateUpdated notification, and unchanged final read revision; it also caught and fixed reversed latest-record ordering in map_epiphany_scene."}
{"ts": "2026-04-26T01:08:00+00:00", "type": "verification", "status": "ok", "note": "Phase 6 jobs reflection landed as read-only thread/epiphany/jobs and live-smoked through app-server with tools/epiphany_phase6_jobs_smoke.py. The surface derives retrieval-index, graph-remap, verification, and specialist slots from thread state/retrieval summaries, preserves revision identity, emits no stateUpdated notification, and leaves live scheduling/progress notifications as future work."}
{"ts": "2026-04-26T01:58:10+00:00", "type": "verification", "status": "ok", "note": "Phase 6 targeted context reflection landed as read-only thread/epiphany/context and live-smoked through app-server with tools/epiphany_phase6_context_smoke.py. The surface returns selected graph/evidence shards plus missing ids without mutation, notification, retrieval, proposal, or promotion."}
{"ts": "2026-04-26T02:19:20+00:00", "type": "decision", "status": "ok", "note": "Context pressure must become a real runtime signal for automatic CRRC. Current agents can infer pressure from transcript length, warnings, quota pressure, and shrinking safe scope, but that is a heuristic scar rather than telemetry."}
{"ts": "2026-04-26T02:39:00+00:00", "type": "research", "status": "ok", "note": "Context-pressure source trace found the correct runtime seam: core TokenUsageInfo already carries total/last usage and model_context_window, core automatic compaction already derives its trigger from ModelInfo::auto_compact_token_limit(), and app-server already emits/replays thread/tokenUsage/updated. The next slice should expose read-only pressure telemetry from those sources, not invent a scheduler or vibes gauge."}
{"ts": "2026-04-26T02:48:00+00:00", "type": "decision", "status": "ok", "note": "Compaction during source gathering or slice planning can erase the transient model while the next agent keeps moving from old persisted state. Automatic CRRC needs an in-flight investigation checkpoint, and absent that checkpoint the correct recovery is to re-gather before implementation rather than continue from false continuity."}
{"ts": "2026-04-26T04:05:00+00:00", "type": "verification", "status": "ok", "note": "Phase 6 context-pressure reflection landed as read-only thread/epiphany/pressure. Core token telemetry now carries model_auto_compact_token_limit beside model_context_window; app-server derives unknown/low/elevated/high/critical pressure from live or stored token usage without compaction, state mutation, scheduling, or stateUpdated notification. Focused protocol/app-server/core checks, schema fixture checks, app-server build, codex-tui check, pressure smoke, and scene smoke passed."}
{"ts": "2026-04-26T04:11:04+00:00", "type": "verification", "status": "ok", "note": "Phase 6 durable investigation checkpointing landed as authoritative typed state, writable only through thread/epiphany/update or accepted promote, rendered into the prompt, reflected through scene/context, and validated against known evidence ids. Focused epiphany-core, codex-core, protocol, and app-server tests plus phase5/scene/context smokes passed."}